DeepSeek V4 Flash Forces Competitors to Slash API Prices by 80%
The release and aggressive pricing of DeepSeek V4 Flash, a sparse Mixture-of-Experts (MoE) model with 284 billion total and 13 billion active parameters, has pressured API providers to slash their prices by up to 80%. This disruption highlights an ongoing price war in the LLM market, where highly efficient open-weights models make proprietary APIs financially unviable unless they drastically lower costs. DeepSeek V4 Flash features a 1-million token context window and is optimized for coding, reasoning, and agent workflows. Despite its massive 284B size, its MoE architecture only activates 13B parameters per token, allowing for highly efficient inference.
## BACKGROUND
An open-weights model provides the trained parameters to the public, allowing developers to run, fine-tune, or quantize the model locally. Mixture-of-Experts (MoE) is an architecture that routes inputs to specific "expert" sub-networks, enabling large models to achieve high performance with only a fraction of their parameters active during computation.